Append options for floating-point commutivity when related diagnostics are produced.

With this patch clang appends the command line options that would allow vectorization when floating-point commutativity is required. Specifically those are enabling fast-math or specifying a loop hint. 

llvm-svn: 244492